Excerpts on the Euler-Maclaurin summation formula, from

Institutiones Calculi Differentialis by Leonhard Euler.

Translated by David Pengelley
c 2000 by David Pengelley
Copyright
(individual educational use only may be made without permission)
Leonhard Eulers book Institutiones Calculi Differentialis (Foundations
of Differential Calculus) was published in two parts in 1755 [2, series I, vol.
10], and translated into German in 1790 [3]. There is an English translation
of part 1 [4], but not part 2, of the Institutiones.
Euler was entranced by infinite series, and a wizard at working with them.
A lot of the book is actually devoted to the relationship between differential
calculus and infinite series, and in this respect it differs considerably from
todays calculus books. In chapters 5 and 6 of part 2 Euler presents his way
of finding sums of series, both finite and infinite, via his discovery of the
Euler-Maclaurin summation formula.
Here we present translated excerpts from these two chapters of part 2,
featuring aspects of Eulers development and applications of his summation
formula. The English translation has been made primarily from the German
translation of 1790 [3], with some assistance from Daniel Otero in comparing
the resulting English with the Latin original. The excerpts include those in
a forthcoming book of annotated original sources, within a chapter on The
Bridge Between the Continuous and Discrete. The book chapter follows
this theme via sources by Archimedes, Fermat, Pascal, Bernoulli and Euler.
For more information see http://math.nmsu.edu/~history. In the spirit
of providing pure uninterpreted translation, we have here removed all our
annotation and commentary, which along with extensive exercises can be
found in the book chapter. The version with annotation and exercises may
be provided at this site at a later time. The only commentary that remains
here summarizes some of those portions of Chapters 5 and 6 that were not
translated.
In excerpts from chapter 5 we see Euler derive his summation formula,
analyze the nature of its Bernoulli numbers in connection with trigonometric
functions, find the precise sums of infinite series of reciprocal even powers,
and prove Bernoullis sums of powers formulas. From chapter 6 we see three
diverse applications of the summation formula, each revealing a fundamentally different way of using it. We first see Euler
large partial
P approximate
1
sums of the slowly diverging harmonic series i=1 i , which involves approximating the now famous Euler constant. Then we see how in the early
1730s Euler approximated the infinite sum of reciprocal squares to great
precision without knowledge of the infinite sum itself. Finally Euler goes on
to use the summation formula to study sums of logarithms, from which he
obtains incredibly impressive formulas and approximations for large factorials (Stirlings series), and thence for binomial coefficients, using Walliss
formula for to determine the unknown constant in his summation formula.

Leonhard Euler, from

Foundations of Differential Calculus
Part Two, Chapter 5
On Finding Sums of Series from the General Term
103. Suppose y is the general term of a series, belonging to the index x,
and thus y is any function of x. Further, suppose Sy is the summative term of
this series, expressing the aggregate of all terms from the first or another fixed
term up to y, inclusive. The sums of the series are calculated from the first
term, so that if x = 1, y is the first term, and likewise Sy yields this first term;
alternatively, if x = 0, the summative term Sy vanishes, because no terms are
being summed. With these stipulations, the summative term Sy is a function
of x that vanishes if one sets x = 0.
[...]
105. Consider a series whose general term, belonging to the index x, is y,
and whose preceding term, with index x 1, is v; because v arises from y, when
x is replaced by x 1, one has
dy
ddy
d3 y
d4 y
d5 y
+

+ etc.
dx 2dx2 6dx3 24dx4 120dx5
If y is the general term of the series
v=y

1 2 3 4 x 1 x
a+ b + c+d+ + v +y
and if the term belonging to the index 0 is A, then v, as a function of x, is the
general term of the series
1 2 3 4 5 x
A+a+ b+ c+d+ + v ,
so if Sv denotes the sum of this series, then Sv = Sy y + A. If one sets
x = 0, then Sy = 0 and y = A, so Sv vanishes.
106. Because
dy
ddy
d3 y
v=y
+

+ etc.,
dx 2dx2 6dx3
one has, from the preceding,
ddy
d3 y
d4 y
dy
+S

S
+
S
etc.,
dx
2dx2
6dx3
24dx4
and, because Sv = Sy y + A,
Sv = Sy S

yA=S

dy
ddy
d3 y
d4 y
S
+
S

S
+ etc.,
dx
2dx2
6dx3
24dx4

or equivalently
S

ddy
d3 y
d4 y
dy
=yA+S

S
+
S
etc.
dx
2dx2
6dx3
24dx4

3
3

ddy d y d y
Thus if one knows the sums of the series, whose general terms are dx
2 , dx3 , dx4 ,
dy
etc., one can obtain the summative term of the series whose general term is dx
.
dy
disappears
The constant A must then be such that the summative term S dx
when x = 0, and this condition makes it easier to determine, than saying that
it is the term belonging to the index 0 in the series whose general term is y.

In 107/108 Euler illustrates the practical application of this equation by

choosing to use the power function y = xn+1 /(n+1). This has the advantage
that the derivatives in the equation are just lower power functions, so that
the sums are all sums of powers, and then vanish after somePpoint in the
equation, so he obtains a finite expression for Sxn (i.e., for xi=1 in ). He
applies this inductively from n = 0 upwards to calculate the closed formulae
for sums of powers of the natural numbers explicitly up through the sum of
fourth powers.
[...]
109. Since from the above one has
dy
1 ddy 1 d3 y
1 d4 y
1 d5 y
S
= y [A] + S 2 S 3 + S 4
S
+ etc.,
dx
2 dx
6 dx
24 dx
120 dx5
3

ddy
dy
dz d y
ddz
= z, then dx
if one sets dx
2 = dx , dx3 = dx2 , etc. And because dy = zdx, y will
R
be a quantity whose differential is zdx, and this one writes as y = zdx. Now
the determination of the quantity y from z according to this formula assumes
R
the integral calculus; but we can nevertheless make use of this expression zdx,
if for z we use no function other than that whose differential is zdx from above.
Thus substituting these values yields
Z
1 dz
1 ddz
1 d3 z
Sz = zdx + S
S 2 + S 3 etc.,
2 dx 6 dx
24 dx
adding to it a constant value such that when x = 0, the sum Sz also vanishes.
110. But if in the expressions above one substitutes the letter z in place of
y, or if one differentiates the preceding equation, which yields the same, one
obtains
dz
1 ddz 1 d3 z
1 d4 z
S
= z + S 2 S 3 + S 4 etc.;
dx
2 dx
6 dx
24 dx
dz
but using dx in place of y one obtains

dz
1 d3 z
1 d4 z
1 d5 z
ddz
=
+
S

S
+
S
etc.
dx2
dx 2 dx3 6 dx4 24 dx5

Similarly replacing y successively by the values

ddz d3 z
,
dx2 dx3

etc., produces

d3 z
ddz 1 d4 z
1 d5 z
1 d6 z
=
+
S

S
+
S
etc.,
dx3
dx2 2 dx4 6 dx5 24 dx6
d3 z
1 d5 z
1 d6 z
1 d7 z
d4 z
S 4 = 3 + S 5 S 6 + S 7 etc.,
dx
dx
2 dx
6 dx
24 dx
and so forth indefinitely.
S

4
3

dz
d z
ddz
111. Now when these values for S dx
, S dx
2 , S dx3 are successively substituted
in the expression
Z
1 dz
1 ddz
1 d3 z
Sz = zdx + S
S 2 + S 3 etc.,
2 dx 6 dx
24 dx
R
dz ddz d3 z
one finds an expression for Sz, composed of the terms zdx, z, dx
, dx2 , dx3
etc., whose coefficients are easily obtained as follows. One sets
Z
dz ddz d3 z d4 z
+
+
+ etc.,
Sz = zdx + z +
+
dx
dx2
dx3
dx4

and substitutes for these terms the values they have from the previous series,
yielding
R
dz
ddz
1
d3 z
1
d4 z
zdx = Sz 21 S dx
+ 16 S dx
2 24 S dx3 + 120 S dx4 etc.
z =
dz
dx
ddz
dx2
d3 z
dx3

dz
ddz
+ S dx
2 S dx
2 +

d3 z
6 S dx3

d4 z
24 S dz 4

+ etc.

ddz
S dx
2

d3 z
2 S dx3

d4 z
6 S dx4

etc.

2S

d4 z
dx4

+ etc.

d4 z
dx4

etc.

d3 z
dx3

etc.
Since these values, added together, must produce Sz, the coefficients , , ,
etc. are defined by the sequence of equations
1
1

1
= 0, + = 0, +
= 0,
2
2
6
2
6
24

1
+
+
= 0, +
+

= 0,
2
6
24 120
2 6 24 120 720

1
+
+

+
= 0 etc.
2 6 24 120 720 5040

112. So from these equations the successive values of all the letters , , ,
etc. are defined; they are
1
1
1

1
= , = = , = +
= 0,
2
2
6
12
2
6
24

1
1

1
= +

=
, = +

+
= 0 etc.,
2
6 24 120
720
2 6 24 120 720
and if one continues in this fashion one finds that alternating terms vanish. The
third, fifth, seventh letters, and so on, in fact all odd terms except the first, are
zero, so that this series appears to contradict the law of continuity by which the
terms proceed. A rigorous proof is especially needed that all odd terms except
the first vanish.

113. Because the letters are determined from the preceding by a constant
law, they form a recurrent series. In order to develop this, consider the series
1 + u + u2 + u3 + u4 + u5 + u6 + etc.,
and set its value = V , so it is clear that this recurrent series arises from the
development of the fraction
1
V =
1
1 2
1 3
1 4
1 2 u + 6 u 24 u + 120
u etc.
And when this fraction is resolved in a different way in an infinite series according
to the powers of u, then necessarily the same series
V = 1 + u + u2 + u3 + u4 + u5 + etc.
will always result. In this fashion a different rule for determining the letters ,
, , etc. results.
114. Because one has
1
1
1 5
1
u + etc.,
eu = 1 u + u2 u3 + u4
2
6
24
120
where e denotes the number whose hyperbolic logarithm is one, then
1 eu
1
1
1
1 4
= 1 u + u2 u3 +
u etc.,
u
2
6
24
120
and thus

u
.
1 eu
Now one removes from this series the second term u = 12 u, so that
V =

1
V u = 1 + u2 + u3 + u4 + u5 + u6 + etc.;
2
whence
1
u (1 + eu )
1
V u= 2
.
2
1 eu
1

Multiplying numerator and denominator by e 2 u yields

1

u
12 u
2
u
e
+
e
1
,
V u= 1
1
2
2 e 2 u e 2 u
1

and converting the quantities e 2 u and e 2 u into series gives

2

u
u
u
1 + 24
+ 2468
+ 24681012
+ etc.
1

V u=
2
4
u
u
2
2 12 + 246
+ 246810
+ etc.

or
2

u
u
u
u
1 + 24
+ 2468
+ 2412
+ 2416
+ etc.
1
V u=
4
6
8
2
u
u
u
u
2
1 + 46 + 46810 + 4614 + 4618 + etc.

115. Since no odd powers occur in this fraction, likewise none can occur in
its expansion; because V 12 u equals the series
1 + u2 + u3 + u4 + u5 + u6 + etc.,
the coefficients of the odd powers , , , etc. all vanish. And so it is
clear why the even-ordered terms after the second all equal zero in the series
1 + u + u2 + u3 + u4 + etc., for otherwise the law of continuity would be
violated. Thus
1
V = 1 + u + u2 + u4 + u6 + u8 + u10 + etc.,
2
and if the letters , , , , have been determined by the development of the
above fraction, one obtains the summative term Sz of the series, whose general
term = z corresponds to the index x, expressed as
Z
1
dz d3 z d5 z d7 z
Sz = zdx + z +
+
+
+ etc.
+
2
dx
dx3
dx5
dx7
116. Since the series 1 + u2 + u4 + u6 + u8 + etc. arises by developing
the fraction
u2
u4
24 + 2468
u2
u4
46 + 46810

1+
1+

+
+

u6
24681012 + etc.
,
u6
468101214 + etc.

the letters , , , , x will follow according to the rule, as

=
=
=
=

1
24
1
2468
1
24612
1
24616

1
46

1
46
46810

1
46
46810
4614

1
46
46810
4614
4618

etc.
But these values are alternatingly positive and negative.
117. If the letters are alternatingly taken negatively, so that
Z
1
dz d3 z d5 z d7 z
Sz = zdx + z
+

+
etc.,
2
dx
dx3
dx5
dx7
then the letters , , , , etc. are determined by the fraction
1
1

u2
u4
u6
u8
24 + 2468 2412 + 2416 etc.
,
u2
u4
u6
u8
+

etc.
46
46810
4614
4618

when one develops it to the series

1 + u2 + u4 + u6 + u8 + etc.

From this one has

1
1

46 24

1
1
=

+
4 6 4 6 8 10 2 4 6 8

1
1
=

4 6 4 6 8 10 4 6 14 2 4 12
etc.;

only here all terms are negative.

118. Thus we consider = A; = B; = C, etc.; consequently
Z
1
Adz Bd3 z Cd5 z Dd7 z
Sz = zdx + z +
+

+ etc.,

2
dx
dx3
dx5
dx7
and in order to determine the letters A, B, C, D etc., we consider the series
1 Au2 Bu4 Cu6 Du8 Eu10 etc.,
which arises from the development of the fraction
1
1

u2
u4
u6
u8
24 + 2468 2412 + 2416 etc.
,
u2
u4
u6
u8
46 + 46810 4614 + 4618 etc.

or consider the series

1
Au Bu3 Cu5 Du7 Eu9 etc. = s,
u
which arises from the development of the fraction
s=

1
u

u2
u4
u6
24 + 2468 2412 + etc.
.
u3
u5
u7
+

+
etc.
46
46810
4614

But since
u2
u4
u6
1
+

+ etc.,
cos u = 1
2
2 4 2 4 6 8 2 4 12
1
u
u3
u5
u7
sin u =
+

+ etc.,
2
2 2 4 6 2 4 6 8 10 2 4 14
we have
s=

cos 12 u
1
1
1 = 2 cot 2 u.
2 sin 2 u

Thus if one converts the cotangent of the arc 12 u into a series, according to the
powers of u, the values of the letters A, B, C, D, E, etc. are revealed.
119. Because s = 21 cot 21 u, one has 12 u = A. cot 2s, and if one differentiates,
2ds
, or 4ds + du + 4ssdu = 0, or
then 21 du = 1+4ss
4ds
+ 1 + 4ss = 0.
du

But since
s=

1
Au Bu3 Cu5 etc.,
u

one has
4ds
du

4
= uu
4A 3 4Bu2 5 4Cu4 7 4Du6 etc.

1=
4ss =

1
4
uu

8A
+

8Bu2

8Cu4

8Du6 etc.

4A2 u2 + 8ABu4 + 8ACu6 + etc.

+ 4BBu6 + etc.

Setting the homogeneous terms to zero, one obtains

A=

1
A2
2AB
2AC + BB
2AD + 2BC
, B=
, C=
, D=
, E=
,
12
5
7
9
11
2AE + 2BD + CC
2AF + 2BE + 2CD
F =
, G=
,
13
15
2AG + 2BF + 2CE + DD
H=
, etc.
17

From these formulas it is very clearly apparent that each of these values is
positive .
120. But because the denominators of these fractions become very large, and
substantially impede calculation, we want instead of the letters A, B, C, D,
etc. to introduce new ones1 :
A=

,
123
D=

B=

,
12345

C=

,
1 2 37

, E=
,
1 2 39
1 2 3 11

etc.

Translators note: Caution! These new symbols , , ... are completely different
from the , , ... used earlier.

Then one finds

1
= ,
2

2
= 2 ,
3

3
= 2 ,
3

4
87 2
= 2 +
,
3
45

5
10 9 8
= 2 + 2
,
3
1 25
=2

=2

12
12 11 10
12 11 10 9 8
+ 2
+
,
123
1 25
1 27

14
14 13 12
14 13 12 11 10
+ 2
+ 2
,
123
1 25
1 27

etc.

121. But it is more convenient to make use of the formulas

4

,
3 2

6
,
3

10
10 9 8
+
,
3
345

12
12 11 10
12 11 10 9 8
+
+

,
3
345
34567
2

1
= ,
2
=

8
8 7 6
+

,
3
345 2

14
14 13 12
14 13 12 11 10
+
+
,
3
345
34567

16
16 15 14
16 15 12
16 15 10
+
+
+

3
345
3 47
3 49
2
etc.

If one finds the values of the letters , , , , etc. according to this rule,
which entails little difficulty in calculation, then one can express the summative
term of any series, whose general term = z corresponding to the index x, in the
following fashion:
Z

dz
d3 z
d5 z
1

+
zdx + z +
2
1 2 3 dx 1 2 3 4 5dx3 1 2 7dx5
7
d z
d9 z
d11 z

+ etc.
1 2 9dx7 1 2 11dx9 1 2 13dx11

Sz =

10

As far as the letters , , , , etc. are concerned, one finds the following
values:

1
2
= 16
= 16
3
= 10
= 56
= 691
210
= 35
2
3617
= 30
= 43867
42
= 1222277
110
854513
= 6
= 1181820455
546
76977927
=
2
= 23749461029
30
8615841276005
=
462

or

1 2 = 1
1 2 3 = 1
1 2 3 4 = 4
1 2 3 5 = 36
1 2 3 6 = 600
1 2 3 7 = 24 691
1 2 3 8 = 20160 35
1 2 3 9 = 12096 3617
1 2 3 10 = 86400 43867
1 2 3 11 = 362880 1222277
1 2 3 12 = 79833600 854513
1 2 3 13 = 11404800 1181820455
1 2 3 14 = 43589145600 76977927
1 2 3 15 = 43589145600 23749461029
1 2 3 16 = 45287424000 8615841276005
etc.

122. These numbers have great use throughout the entire theory of series.
First, one can obtain from them the final terms in the sums of even powers,
for which we noted above (in 63 of part one) that one cannot obtain them, as
one can the other terms, from the sums of earlier powers. For the even powers,
the last terms of the sums are products of x and certain numbers, namely for
1
1
1
, 42
, 30
etc. with alternating signs. But
the 2nd, 4th, 6th, 8th, etc., 16 , 30
these numbers arise from the values of the letters , , , , etc., which we
found earlier, when one divides them by the odd numbers 3, 5, 7, 9, etc. These
numbers are called the Bernoulli numbers after their discoverer Jakob Bernoulli,

11

and they are

11

13

15

17

1
6

1
30
1
42
1
30
5
66

=
=
=

19

=A

{
21 =

=B

23

25

27

29

31

=C
=D
=E

691
= 2730
=F

=
=

7
6
3617
510

=G

43867
798

=
=
=
=
=

174611
330
854513
138
236364091
2730
8553103
6
23749461029
870
8615841276005
14322

=H

= I
= K = 283617
330
= L = 11131593
2323
=M
= N = 13657931
6
=O
=P

etc.

123. Thus one immediately obtains the Bernoulli numbers A, B, C etc. from
the following equations:
A=
B=
C=
D=
E=
F=
G=

43
12
65
12
87
12
109
12
1211
12
1413
12

1
6
1
5
2
7
2
9
2
11
2
13
2
15

A2
AB
AC +
AD +
AE +
AF +

8765
1234
10987
1234
1211109
1234
14131211
1234

19 B2
2
11
BC
2
13
BD +
2
15
BE +

121110987
123456
14131211109
123456

1
13
C2
2
15
CD

etc.,
and the law of these equations is clear if one notes that whenever the square of
a letter appears, its coefficient is only half as large as it would appear according
to the rule. But actually one should view the terms containing products of
different letters as occurring twice. So for example
13F =

12 11
12 11 10 9
12 11 10 9 8 7
AE +
BD +
CC
12
1234
123456

12 11 10 5
12 11 10 3
DB +
EA.
1 2 38
1 2 3 10
124. Next, the numbers , , , etc. are also ingredients in the expressions
for the sums of the series of fractions comprised by the general formula
1
1
1
1
1
1 + n + n + n + n + n + etc.,
2
3
4
5
6
when n is a positive even number. We expressed the sums of these series in
the Introductio 2 via powers of the semiperiphery of the circle of radius = 1,
and there one encounters the numbers , , , , etc. in the coefficients of
+

Introductio, Book I, chapter 10.

12

these powers. But because these do not appear to occur by accident, rather
their necessity is apparent, we wish to investigate these sums in a special way,
by which the truth of the law of these sums will be clear. Because from above
(43) one has

m
1
1
1
1
1
1
cot =

+ etc.,
n
n
m n m n + m 2n m 2n + m 3n m
combining terms in pairs one has,

2m
2m
2m
m
1
2m
cot =

2
2

etc.,
2
2
2
n
n
m nn m
4n m
9n m
16n2 m2
and from this
1
1
1
1
1

m
+
+
+
+ etc. =

cot .
n2 m2 4n2 m2 9n2 m2 16n2 m2
2mm 2mn
n
Now we set n = 1 and replace m by u, yielding
1

1
1
1
1

cot u.
+
+
+
+ etc. =
1 u2 4 u2 9 u2 16 u2
2uu 2u
Resolving these fractions in series, one obtains
1
1u2

= 1 + u2 + u4 + u6 + u8 + etc.

1
4u2

= 212 + u24 + u26 + u28 + 2u10 + etc.

1
9u2

= 312 + u34 + u36 + u38 + 3u10 + etc.

1
16u2

= 412 + u44 + u46 + u48 + 4u10 + etc.

125. If we thus set

1 + 212 + 312 + 412 + etc. = a

1+

1 + 214
1 + 216

+ etc. = b

1+

+ etc. = c

1+

+ 314
+ 316

+ 414
+ 416

1
28
1
210
1
212

+
+
+

1
38
1
310
1
312

+
+
+

1
48
1
410
1
412

+ etc. = d
+ etc. = e
+ etc. = f

etc.,
then the series above is transformed into

cot u.
2uu 2u
Now in 118 we found that for the letters A, B, C, D etc., when one sets
a + bu2 + cu4 + du6 + eu8 + fu10 + etc. =

s=

1
Au Bu3 Cu5 Du7 Eu9 etc.
u

13

one has s = 21 cot 12 u, and thus, when one replaces 12 u by u, or u by 2u, one
obtains
1
1
cot u =
2Au 23 B 3 u3 25 C 5 u5 27 D 7 u7 etc.,
2
2u
and multiplying by u yields

1
cot u =
2A 2 23 B 4 u2 25 C 6 u4 27 D 8 u6 etc.,
2u
2uu
from which follows
1

cot u = 2A 2 + 23 B 4 u2 + 25 C 6 u4 + 27 D 8 u6 + etc.
2uu 2u
Since we already found that
1

cot u = a + bu2 + cu4 + du6 + etc.,

2uu 2u
it necessarily follows that
2
123

2 =

2A
12

b = 23 B 4 =

23
12345

4 =

23 B
1234

c = 25 C 6 =

25
1237

6 =

25 C
126

d = 27 D 8 =

27
1239

8 =

27 D
128

a=2

A 2 =

11

11

2
2 E
e = 29 E 10 = 12311
10 = 1210
10
2
2 F
f = 211 F 12 = 12313
12 = 1212
12

etc.
[...]
129. From the table of values of the numbers , , , etc. that we
communicated above in 121, it is apparent that they at first decrease, but
then grow without end. Thus it is worth the effort to investigate in what ratio
these numbers continue to grow, after they reach considerable size. So let be
a number far from the beginning in the sequence , , , , etc., and the one
immediately following. Since the sums of the reciprocal powers are determined
by these numbers, we let 2n be the exponent of the power, in whose sum
occurs; 2n + 2 will be the exponent of the power corresponding to , and n a
very large number. Then from 125 one has
1
1
1
22n1
+
+
+
etc.
=
2n ,
22n 32n 42n
1 2 3 (2n + 1)
1
1
1
22n+1
1 + 2n+2 + 2n+2 + 2n+2 + etc. =
2n+2 .
2
3
4
1 2 3 (2n + 3)
Dividing this series by the former, one finds
1+

1+
1

1
22n+2
1
+ 22n

+
+

1
+ etc.
32n+2
1
+ etc.
32n

4 2
.
(2n + 2) (2n + 3)

14

But because n is a very large number and both series are very closely = 1,

(2n + 2) (2n + 3)
nn
=
=
.
2

Now n indicates which term the number is beyond the first number , and
from this the number is to the following as 2 is to n2 , and this ratio
would, if n were an infinitely large number, be in complete accordance with the
truth. Because nearly = 10, when one lets n = 100, the hundredth term is
approximately 1000 times smaller than the subsequent one. Thus the numbers
, , , etc., and also the Bernoullian A, B, C, D etc., form a highly diverging
sequence, which grows more strongly than any geometric sequence of growing
terms.
130. Thus if one has found the numbers , , , etc., or A, B, C, D
etc., then given a series, whose general term z is a function of its index x, the
summative term Sz can be expressed as follows:
Z

1
1
dz
1
d3 z
zdx + z +

2
6 1 2dx 30 1 2 3 4dx3
5
d z
d7 z
1
1

+
42 1 2 3 6dx5 30 1 2 3 8dx7
d9 z
d11 z
5
691

66 1 2 3 10dx9 2730 1 2 3 12dx11

7
d13 z
d15 z
3617
+

6 1 2 3 14dx13
510 1 2 3 16dx15
d17 z
174611
d19 z
43867

+
17
798 1 2 3 18dx
330
1 2 3 20dx19
854513
d21 z
d23 z
236364091
+

21
138
1 2 3 22dx
2730
1 2 3 24dx23
8553103
d25 z
23749461029
d27 z
+

6
1 2 3 26dx25
870
1 2 3 28dx27
29
8615841276005
d z
+

etc.
14322
1 2 3 30dx29
R
Thus if one knows the integral zdx, or the quantity, whose differential is
= zdx, one finds the summative term by means of continuing differentiation.
One must not neglect that a constant value must always be added to this
expression, of a nature that the sum will = 0, when x becomes 0.
Sz =

131. If now z is an integral rational function of x, so that the derivatives

eventually vanish, then the summative term is represented by a finite expression.
We illustrate this by some examples.
First example.
Find the summative term of the following series.

15

1 + 9 + 25 + 49 + 81 + + (2x 1)2
Since here z = (2x 1)2 = 4xx 4x + 1, one has
Z

4
zdx = x3 2x2 + x;
3

because from this, differentiation produces 4xxdx 4xdx + dx = zdx. Further

differentiation yields
dz
= 8x 4,
dx

ddz
= 8,
dx2

d3 z
= 0 etc.
dx3

So the summative term sought equals

1 2
1
4 3
x 2x2 + x + 2xx 2x + + x Const.,
3
2 3
3
in which the constant must remove the terms

1
2

13 , so

4
1
x
S (2x 1)2 = x3 x = (2x 1) (2x + 1) .
3
3
3
So if one sets x = 4, the sum of the first four terms
1 + 9 + 25 + 49 =

4
7 9 = 84.
3

[...]
132. From this general expression for the summative term, the sum for powers
of natural numbers, that we communicated in the first part (29 and 61), but
which we
at that time, follows very easily. Let us set z = xn ,
R could not1 prove
n+1
, and differentiating,
so that zdx = n+1 x
3
dz
ddz
n2 d z
= nxn1 ,
=
n
(n

1)
x
,
= n (n 1) (n 2) xn3 ,
dx
dx2
dx3
7
d5 z
n5 d z
=
n
(n

1)
(n

2)
(n

3)
(n

4)
x
,
= n (n 1) (n 6) xn7 , etc.
dx5
dx7

16

From this we deduce the following summative term corresponding to the general
term xn :
1
1 n
1 n (n 1) (n 2) n3
1
xn+1 + xn + xn1

x
Sxn =
n+1
2
6 2
30
234
1 n (n 1) (n 2) (n 3) (n 4) n5
+
x
42
23456
1 n (n 1) (n 6) n7

x
30
2 38
5 n (n 1) (n 8) n9
+
x
66
2 3 10
691 n (n 1) (n 10) n11

2730
2 3 12
7 n (n 1) (n 12) n13
+
x
6
2 3 14
3617 n (n 1) (n 14) n15

510
2 3 16
43867 n (n 1) (n 16) n17

x
+
798
2 3 18
174611 n (n 1) (n 18) n19

x
330
2 3 20
854513 n (n 1) (n 20) n21
+

x
138
2 3 22
236364091 n (n 1) (n 22) n23

x
2730
2 3 24
8553103 n (n 1) (n 24) n25
+

x
6
2 3 . . . 26
23749461029 n (n 1) (n 26) n27

x
870
2 3 28
8615841276005 n (n 1) (n 28) n29
+

x
14322
2 3 30
etc.
This expression differs from the former only in that here we have introduced
the Bernoulli numbers A, B, C, D etc., whereas above we used the numbers
, , , etc.; the agreement is clear. Thus here we have been able to give
the summative terms for all powers up to the thirtieth, inclusive; if we wanted
to perform this investigation via other means, lengthy and tedious calculations
would be necessary.
[...]
Part Two, Chapter 6
On the summing of progressions via infinite series
140. The general expression, that we found in the previous chapter for the
summative term of a series, whose general term corresponding to the index x

17

is z, namely
Z
Sz =

1
Adz
Bd3 z
Cd5 z
zdx + z +

+
etc.,
2
1 2dx 1 2 3 4dx3 1 2 6dx5

actually serves to determine the sums of series, whose general terms are integral
rational functions of the index x, because in these cases one eventually arrives
at vanishing differentials. On the other hand, if z is not such a function of x,
then the differentials continue without end, and there results an infinite series
that expresses the sum of the given series up to and including the term whose
index = x. The sum of the series, continuing without end, is thus given by
taking x = , and one finds in this way another infinite series equal to the
original.
141. If one sets x = 0, the expression represented by the series must vanish,
as we already noted; and if this does not occur, one must add to or take away
from the sum a constant amount, so that this requirement is satisfied. If this is
the case, then when x = 1 one obtains the first term of the series, when x = 2
the sum of the first and second, when x = 3 the sum of the first three terms of
the series, etc. Because in these cases the sum of the first, first two, first three,
etc. terms is known, this is also the value of the infinite series expressing the
sum; and thus one is placed in a position to sum countlessly many series.
142. Since when a constant value is added to the sum, so that it vanishes
when x = 0, the true sum is then found when x is any other number, then it
is clear that the true sum must likewise be given, whenever a constant value
is added that produces the true sum in any particular case. Thus suppose it
is not obvious, when one sets x = 0, what value the sum assumes and thus
what constant must be used; one can substitute other values for x, and through
addition of a constant value obtain a complete expression for the sum. Much
will become clear from the following.
142a. Consider first the harmonic progression
1+

1 1 1
1
+ + + + = s.
2 3 4
x

Since the general term = x1 , we have z = x1 , and the summative term s will
R
R
be found as follows. First one has zdx = dx
x = lx; from differentiation one
has
dz
1 ddz
1
d3 z
1
d4 z
1
d5 z
1
= 2,
=
,
=

,
=
,
= 6 , etc.
dx
x 2dx2
x3 6dx3
x4 24dx4
x5 120dx5
x
From this
s = lx +

1
A
B
C
D
2 + 4 6 + 8 etc. + Constant.
2x 2x
4x
6x
8x

18

However, the added constant value cannot be determined from the case when
x = 0. So we set x = 1. Since then s = 1, one has
1=

1 A B C D
+
+ etc. + Constant,
2
2
4
6
8

and thus the constant is

=

1 A B C D
+
+ + etc.
2
2
4
6
8

Consequently the summative term sought is

B
C
D
1
s = lx + 2x
2xA2 + 4x
4 6x6 + 8x8 etc.

1
2

A
2

B
4

C
6

D
8

+ etc.

143. Since the Bernoulli numbers A, B, C, D etc. form a diverging series, it

is not possible to really know the value of the constant here. But if we substitute
a larger number for x, and really find the sum of that many terms, the value of
the constant can be found easily. Let us set as the end x = 10; the sum of the
first ten terms
= 2, 928968253968253968,
which must equal the expression for the sum when one sets x = 10 in it, yielding
l10 +

A
B
C
D
1

+
etc. + C.
20 200 40000 6000000 800000000

Thus if one substitutes for l10 the hyperbolic logarithm of 10, and in place of
A, B, C etc. substitutes the values found above [122], one obtains for the
constant
C = 0, 5772156649015325,
and this number therefore expresses the sum of the series
1 A B C D
E
+
+ +
etc.
2
2
4
6
8
10
144. If one substitutes for x a not very large number, then the sum of the
[original] series is easy to find, and one obtains the sum
1
A
B
C
D
2 + 4 6 + 8 etc. = s lx C.
2x 2x
4x
6x
8x
But if x is a very large number, then the sum of this infinite expression can be
found in decimal fractions. Now it is clear to begin with that if the [original]
series continues infinitely, the sum will have infinite magnitude, because as
x = , also lx grows to infinity. But in order nonetheless to be able to give the
sum of any number of terms more easily, we express the values of the letters A,

19

B, C etc. in decimal fractions.

A = 0, 1666666666666
B = 0, 0333333333333
C = 0, 0238095238095
D = 0, 0333333333333
E = 0, 0757575757575
F = 0, 2531135531135
G = 1, 1666666666666
H = 7, 0921568627451 etc.,

and so

A
2
B
4
C
6
D
8
E
10
F
12
G
14
H
16

= 0, 0833333333333
= 0, 0083333333333
= 0, 0039682539682
= 0, 0041666666666
= 0, 0075757575757
= 0, 0210927960928
= 0, 0833333333333
= 0, 4432598039216 etc.

First example.
Find the sum of one thousand terms of the series 1 +

1
2

1
3

1
4

+ 15 + etc.

20

Set x = 1000, and because

l10 =
one has lx =
Const. =
1
=
2x

2, 3025850929940456840,
6, 9077552789821
0, 5772156649015
0, 0005000000000
7, 4854709438836

A
= 0, 0000000833333
2xx
7, 4854708605503
B
add
= 0, 0000000000000,
4x4
thus
7, 4854708605503
subtr.

is the desired sum of a thousand terms, which is still not seven and a half.
[...]
148. After considering the harmonic series we wish to turn to examining the
series of reciprocals of the squares, letting
1 1
1
1
+ +
+ +
.
4 9 16
xx
R
1
Since the general term of this series is z = xx
, then zdx = x1 , the differentials
of z are
s=1+

dz
1
= 3,
2dx
x

ddz
1
= 4,
2 3dx2
x

d3 z
1
= 5
2 3 4dx3
x

etc.,

and the sum is

s=C

1
1
A
B
C
D
E
+

+ etc.,
x 2xx x3 x5 x7 x9 x11

where the added constant C is determined from one case in which the sum is
known. We therefore wish to set x = 1. Since then s = 1, one has
C =1+1

1
+ A B + C D + E etc.,
2

but this series alone does not give the value of C, since it diverges strongly.
Above [125] we demonstrated that the sum of the series to infinity is =
6 ,

and therefore setting x = , and s =

,
we
have
C
=
,
because
then
all
6
6
other terms vanish. Thus it follows that

1
.
1 + 1 + A B + C D + Eetc. =
2
6
149. If the sum of this series were not known, then one would need to
determine the value of the constant C from another case, in which the sum

21

were actually found. To this aim we set x = 10 and actually add up ten terms,
obtaining
s

= 1, 549767731166540690.

subtr.

1
x
1
2xx

add

A
x3

= 0, 000166666666666666
1, 644934397833207356

subtr.

B
x5

= 0, 000000333333333333
1, 644934064499874023

add

C
x7

= 0, 000000002380952381
1, 644934066880826404

subtr.

D
x9

add

E
x11

= 0, 000000000000757575
1, 644934066848250646

subtr.

F
x13

= 0, 000000000000025311
1, 644934066848225335

add

G
x15
H
x17

= 0, 000000000000001166

Further, add

subtr.

= 0, 1
= 0, 005
1, 644767731166540690

0, 000000000033333333
1, 644934066847493071

71
1, 644934066848226430 = C.

This number is likewise the value of the expression

6 , as one can find by
calculation from the known value of . From this it is clear that, although the
series A, B, C, etc. diverges, it nevertheless produces a true sum. [...]
In 150153 Euler explores possible formulas for exact sums of the infinite
series of reciprocal odd powers of the natural numbers, similar to those he
has already found for the sums of reciprocal even powers in terms of the
Bernoulli numbers and . Using his summation formula he produces highly
accurate decimal approximations for the sums of reciprocal odd powers all
the way through the fifteenth power, hoping to see a pattern analogous to
the even powers, namely simple fractions times the relevant power of . He
is disappointed, however, not to find that they behave similarly to the even
powers.
Then in 154156 Euler uses a sum and the inverse tangent and cotangent
functions to approximate to seventeen decimal places with his summation
formula, and remarks that it is amazing that one can approximate so
accurately with such an easy calculation.

22

157. Now we want to use for z transcendental functions of x, and take

z = lx for summing hyperbolic logarithms, from which the ordinary can easily
be recovered, so that
s = l1 + l2 + l3 + l4 + + lx.
Because z = lx,

Z
zdx = xlx x,

since its differential is dxlx. Then

dz
1 ddz
1
d3 z
1
d4 z
1
d5 z
1
= ,
=

,
=
,
=

,
= 5 , etc.
2
2
3
3
4
4
5
dx
x dx
x 1 2dx
x 1 2 3dx
x 1 2 3 4dx
x
One concludes that
A
B
C
D
1

+ etc. + Const.
s = xlx x + lx +
3
5
2
1 2x 3 4x
5 6x
7 8x7
But for this constant one finds, when one sets x = 1, because then s = l1 = 0,
B
C
D
A
+

+
etc.,
C =1
12 34 56 78
a series that, due to its great divergence, is quite unsuitable even for determining
the approximate value of C.
158. Nevertheless we can not only approximate the correct value of C, but
can obtain it exactly, by considering Walliss expression for provided in the
Introductio [1, vol. 1, chap. 11]. This expression is

2 2 4 4 6 6 8 8 10 10 12 etc.
=
2
1 3 3 5 5 7 7 9 9 11 11 etc.
Taking logarithms, one obtains from this
l l2 = 2l2 + 2l4 + 2l6 + 2l8 + 2l10 + l12 + etc.
l1 2l3 2l5 2l7 2l9 2l11 etc.
Setting x = in the assumed series, we have

l1 + l2 + l3 + l4 + + lx = C + x + 21 lx x,

thus
l1 + l2 + l3 + l4 + + l2x = C + 2x + 21 l2x 2x

and
l2 + l4 + l6 + l8 + + l2x = C + x + 21 lx + xl2 x,

and therefore l1 + l3 + l5 + l7 + + l (2x 1) = xlx + x + 21 l2 x.

Thus because
l 2 = 2l2 + 2l4 + 2l6 + + 2l2x l2x
2l1 2l3 2l5 2l (2x 1) ,

23

letting x = yields

= 2C + (2x + 1) lx + 2xl2 2x l2 lx 2xlx (2x + 1) l2 + 2x,

2
and therefore

1
l = 2C 2l2, thus 2C = l2and C = l2,
2
2
yielding the decimal fraction representation
l

C = 0, 9189385332046727417803297,
thus simultaneously the sum of the series
1

A
B
C
D
E
1
+

+ etc. = l2.
1 2 3 4 5 6 7 8 9 10
2

159. Since we now know the constant C = 12 l2, one can exhibit the sum of
any number of logarithms from the series l1 + l2 + l3+ etc. If one sets
s = l1 + l2 + l3 + l4 + + lx,
then

1
1
A
B
C
D
s = l2 + x +
lx x +

+ etc.
3
5
2
2
1 2x 3 4x
5 6x
7 8x7
if the proposed logarithms are hyperbolic; if however the proposed logarithms
are common, then one must take common logarithms also in the terms 12 l2 +
(x + 12 )lx for l2 and lx, and multiply the remaining terms
B
A

+ etc.
1 2x 3 4x3
of the series by 0, 434294481903251827 = n. In this case the common logarithms are
x +

l = 0, 497149872694133854351268
l2 = 0, 301029995663981195213738
l2 = 0, 798179868358115049565006
1
l2 = 0, 399089934179057524782503.
2
Example.
Find the sum of the first thousand common logarithms
s = l1 + l2 + l3 + + l1000.

24

So x = 1000, and
lx =

3, 0000000000000,

and thus xlx = 3000, 0000000000000

1
2 lx =

1, 5000000000000

1
2 l2 =

0, 3990899341790
3001, 8990899341790

subtr. nx = 434, 2944819032518

2567, 6046080309272.
Then
nA
0, 0000361912068
12x =
subtr.

nB
34x3

0, 0000000000012
0, 0000361912056

add 2567, 6046080309272

the sum sought s = 2567, 6046442221328.
Now because s is the logarithm of a product of numbers
1 2 3 4 5 6 1000,
it is clear that this product, if one actually multiplies it out, consists of 2568
figures, beginning with the figures 4023872, with 2561 subsequent figures.
160. By means of this summation of logarithms, one can approximate the
product of any number of factors, that progress in the order of the natural
numbers. This can be especially helpful for the problem of finding the middle or
largest coefficient of any power in the binomial (a + b)m , where one notes that,
when m is an odd number, one always has two equal middle coefficients, which
taken together produce the middle coefficient of the next even power. Thus
since the largest coefficient of any even power is twice as large as the middle
coefficient of the immediately preceding odd power, it suffices to determine the
middle largest coefficient of an even power. Thus we have m = 2n with middle
coefficient expressed as
2n (2n 1) (2n 2) (2n 3) (n + 1)
.
1 2 3 4n
Setting this = u, one has
u=

1 2 3 4 5 2n
,
(1 2 3 4 n)2

and taking logarithms

lu = l1 + l2 + l3 + l4 + l5 + l2n
2l1 2l2 2l3 2l4 2l5 2ln.

25

161. The sum of hyperbolic logarithms is

1
1
1
l1 + l2 + l3 + l4 + + l2n = l2 + 2n +
ln + 2n +
l2 2n
2
2
2
A
B
C
+

+
etc.
1 2 2n 3 4 23 n3 5 6 25 n5
and
2l1 + 2l2 + 2l3 + 2l4 + + 2ln
2A
2B
2C
= l2 + (2n + 1) ln 2n +

+
etc.
3
1 2n 3 4n
5 6n5
Subtracting this expression from the former yields
1
1
A
B
C
lu = l ln + 2nl2 +

+
etc.
3
3
2
2
1 2 2n 3 4 2 n
5 6 25 n5
2B
2A
2C
+

+ etc.,
1 2n 3 4n3 5 6n5
and collecting terms in pairs

3A
15B
22n
63C
255D

+
lu = l

+
etc.
3
3
5
5
n 1 2 2n 3 4 2 n
562 n
7 8 27 n7
One has
3A
15B
63C
255D

+ etc.
2
2
4
4
6
6
1 2 2 n 3 4 2 n
562 n
7 8 28 n8
A
B
C
D
= l 1 + 2 2 + 4 4 + 6 6 + 8 8 + etc. ,
2 n
2 n
2 n
2 n
so that

22n
A
B
C
lu = l
2nl 1 + 2 2 + 4 4 + 6 6 + etc.
n
2 n
2 n
2 n
and thus
u=
1+

22n
A

22 n2

24 n4

26 n6

2n .
+ etc.
n

26

Setting 2n = m,
l(1 + 22An2 + 24Bn4 + 26Cn6 +
=

A
m2

B
m4
A2
2m4

C
m6
AB
m6

A3
3m6

D
m8
AC
m8
BB
2m8
A2 B
m8

A4
4m8

+
+

D
28 n8
E
m10
AD
m10
BC
m10
A2 C
m10
AB 2
m10
A3 B
m10
A5
5m10

+ etc.)
+ etc.
etc.
etc.
+ etc.
+ etc.
etc.
+ etc.;

and because this expression must equal

3A
15B
63C
255D

+ etc.,
2
4
6
1 2m
3 4m
5 6m
7 8m8
one has
3A
12
A2 15B
B=

2
34
1
63C
C = AB A3 +
3
56
1 2
1
255D
D = AC + B A2 B + A4
2
4
78
A=

1023E
1
E = AD + BC A2 C AB 2 + A3 B A5 +
5
9 10
etc.
162. Now since A = 16 , B =
1
A= ,
4

B=

1
30 ,

1
,
96

C=

C=

1
42 ,

D=

27
,
640

1
30 ,

E=

D=

211

5
66 ,

one has

90031
32 5 7

etc.

Consequently
u=
1+

22n
1
24 n2

1
29 3n4

27
213 5n6

90031
219 32 57n8

2n
n
+ etc.

or
u=

22n 1

1
24 n2

7
29 3n4

121
213 35n6

107489
219 32 57n8

2n
etc.

27

or, if one actually takes the power of the series, approximately

u=
n 1 +

1
4n

1
32n2

22n
1
128n
3

5
16128n4

.
+ etc.

Thus the middle term in (1 + 1)2n is to the sum 22n of all the terms as

1
1
1
1

+ etc. ;
1is to n 1 +
+
4n 32n2 128n3 16 128n4
or, if one abbreviates 4n = , as

1
1
1
5
23
53
1is to n 1 + + 2 3 4 + 5 +
etc. .

2
2
8
8
16 6
[...]
Second Example
Find the ratio of the middle term of the binomial (1 + 1)100 to the sum 2100
of all the terms.
For this we wish to use the formula we found first,
22n
15B
3A
63C
lu = l
+

+ etc.,
n 1 2 2n 3 4 23 n3 5 6 25 n5
from which, setting 2n = m, in order to obtain the power (1 + 1)m , and after
substituting the values of the letters A, B, C, D etc., one has
1
2m
1
1
17
31
691
+
lu = l 1

+
etc.
3
5
7
9
4m 24m
20m
112m
36m
88m11
2 m
Since the logarithms here are hyperbolic, one multiplies by
k = 0, 434294481903251,
in order to change to tables, yielding
2m
k
k
k
17k
31k
lu = l 1

+ etc.,
3
5
7
4m 24m
20m
112m
36m9
2 m
Now since u is the middle coefficient, the ratio sought is 2m : u, and

1
2m
k
k
k
17k
31k
691k
=l
m +

+ etc.
3
5
7
9
u
2
4m 24m
20m
112m
36m
88m11

Now, since the exponent m = 100,

k
= 0, 0043429448,
m

k
= 0, 0000004343,
m3

k
= 0, 0000000000,
m5

28

yielding
k
4m
k
24m3

= 0, 0010857362
= 0, 0000000181
0, 0010857181 .

Further l = 0, 4971498726
l 12 m = 1, 6989700043
l 12 m = 2, 1961198769

l 21 m = 1, 0980599384
k
4m

k
24m3

+ etc. = 0, 0010857181
100

1, 0991456565 = l 2 u .
100

Thus 2 u = 12, 56451, and the middle term in the expanded power (1 + 1)m
is to the sum of all the terms 2100 as 1 is to 12, 56451.
References
[1] L. Euler, Introduction to analysis of the infinite (transl. John D. Blanton), SpringerVerlag, New York, 1988.
[2] L. Euler, Opera Omnia, B.G. Teubner, Leipzig and Berlin, 1911 .
[3] L. Euler, Vollst
andige Anleitung zur Differenzial-Rechnung (transl. Johann Michelsen),
Berlin, 1790, reprint of the 1798 edition by LTR-Verlag, Wiesbaden, 1981.
[4] L. Euler, Foundations of differential calculus (transl. John D. Blanton), Springer Verlag, New York, 2000.